A Model Of Associative Learning In Hermissenda Crassicornis: Sensitin'ities To Temporal Relationships Of Stimuli

Abstract

A computer model of associative learning in the invertebrate, Hermissenda crassicornig, has previously been shown to demonstrate many characteristics of vertebrate conditioning. The model is tested in experimental paradigms which mimic those applied in behavioral studies. Temporal learning characteristics of the model are shown to be quantitatively similar to that of the animal. These results demonstrate the value of this computational learning model as a tool for examining associative learning in biological systems and for uncovering insights concerning associative learning, memory, and recall which have been applied to the development of artificial neural networks. IN'IRODUCI'IONIn [l], we describe a computer model of associative learning as demonstrated in a marine snail, Hermissenda grassicornis [2]. The model aggregates and simplifies the dynamics of the component mechanisms and captures only the essential cellular and network features of associative learning. It reproduces associative learning that is quantitatively similar to that of Hermissenda and exhibits many characteristics of Pavlovian conditioning when the model is trained with paired stimuli. In this paper, we explore sensitivities of the model's learning to the temporal relationship of the conditioning stimulus (CS) and the unconditioned stimulus (US). In particular, we examine the model's sensitivities to interstimulus interval (ISI) and to relative timing of the stimulus offsets. DESCRLFTION OF THE MODELThe studies of associative learning in Hermissenda focus on information flow in a four-neuron circuit that is capable of learning to associate light (CS) and rotation (US), even when isolated from the animal ( Figure 1). The computer model of associative learning incorporates intercellular and intracellular information flow, and physiological and biochemical events essential to the learning process ( Figure 2):1. The resistance-capacitance (RC) circuit concept used to model the currents through (including the generator current dynamics, GCD, in Figure 2) and the voltage across each neuron membrane (membrane dynamics, MD, in Figure 2); 2. Post-inhibitory rebound currents in the E-cell ( a ganglion cell) and vestibular hair cells (rebound current dynamics, RCD, in Figure 2); 3. Voltage-dependent calcium current in the B-cell (calcium current dynamics, CCD, in Figure 2); 4. Calcium-dependent and synapse-specific increase of the B photoreceptor's input resistance and ability to maintain an increased membrane resistance (resistance dynamics, RD, in Figure 2). This synapse-specific mechanism receives an important contribution from GABA-mediated stimulation of the photoreceptor by the caudal hair cell during paired stimulus presentations [3].5. Time-varying shunting of B-cell synaptic inputs during light exposure (RD in Figure 2). The input resistance of the photoreceptor, initially small, gradually increases during a light stimulus [4]. Figure 1 : htenensory integration by the Hermissenda newous system. (A) convergence of synapb;c inhibition trom type ...